RAVEN. / readinessRAVEN. / readinessFourteen questions to answer before wiring Raven into an agent, bot, launchpad, or toolkit — plus a 5-day pilot. Machine-readable: /launchguard-readiness.json.
What action follows Raven (display/list/route/spend/trade/escalate/block)? · What evidence will you pass (mint+program required; metadata/pool optional)? · What may be missing — and which gaps does YOUR policy accept? · Which verdicts require human approval (default: warning/risk/unknowable)? · Staleness window (pre-trade = re-verify immediately; delayed = at execution)? · Where are receipts stored (schema, append-only)? · Who owns policy changes (a named human)? · How are eval failures handled (build orders)? · Verifier down (fail closed, MCP fallback)? · Signature fails (incident)? · Token unsupported (not verified)? · Pilot scope and scale path?
Sync — collect mint, tokenProgram, metadata/pool addresses, action context, policy version. Plan — decide the intended action and which gaps your policy accepts. Analyze — call Raven, verify the signature, inspect verdict/findings/gaps. Deliver — a bounded decision: proceed under policy, block, escalate, or re-verify. Evaluate — store the exact receipt, track false positives/negatives, feed eval fixtures, improve policy. No LLM in the verdict path — this loop is policy, not a planner.
Day 0: integrate the read-only call (no wallet access). Day 1: verify signatures, store receipts. Day 2: apply decision policy. Day 3: run the eval fixtures + public smoke against your wiring. Day 4: review gaps hit and disagreements. Day 5: decide — beta key, expand, or send feedback.
No revenue, performance, safety, or rug-prevention promises — signed evidence and explicit gaps only.